将数据拆分为块并遍历 R 中的每个块

Question

我有一个结构如下的数据框：

birthwt  tobacco01  pscore  pscoreblocks     blocknumber
3425     0          0.18    (0.177, 0.187]   1
3527     1          0.15    (0.158, 0.168]   2
1638     1          0.34    (0.335, 0.345]   3

解释数据：birthwt 列是以克为单位测量出生体重的连续变量。烟草01 列包含值 0 或 1。pscore 列包含介于 0 和 1 之间的概率值。pscoreblocks 采用 pscore 列并将其分解为 100 个大小相同的块。块编号为每个块提供一个编号，因此它从 1 到 100。

我正在尝试对 pscoreblocks 中的每个块执行以下操作。

apply_model <- function(data) {
   one <- lm(birthwt ~ tobacco01, data)
   two <- one$coefficients[[2]]
   two_5 <- ((sum(data$tobacco01 == 1)) + (sum(data$tobacco01 == 0)))/ sum(data$tobacco)
   three <- two*two_5
   return(three)
}

方法 1：一种方法（低效）是使用过滤器为每个块创建单独的数据帧。

data1 <- data %>% filter(blocknumber == 1) 

然后我可以在每个块上手动运行上面的函数。

方法 2：但是，我希望能够高效地运行 100 个块。

已提出以下解决方案here：

data %>% group_split(blocknumber) %>% map(apply_model)

我在这里得到的结果与我使用它时的结果相同：

lapply(split(data, data$blocknumber), apply_model)

问题：

当我将使用方法 1 得到的值与使用方法 2 得到的值进行比较时，我期望得到相同的结果。如果我过滤掉块号 1 并运行分析，而不是查看第二种方法中标记为 (1) 的值，我不会得到相同的值。为什么我在这里没有得到相同的价值？

更一般地说，我如何根据列值将数据拆分成块，然后迭代以运行一个函数，该函数涉及一个指代正在使用的数据帧的术语？

可重现的例子：

> small <- dput(dfcsmall[1:40,])
structure(list(birthwt = c(3629, 3005, 3459, 4520, 3095.17811313023, 
3714, 3515, 3232, 3686, 4281, 2645.29691556227, 3714, 3232, 3374, 
3856, 3997, 3515, 3714, 3459, 3232, 3884, 3235, 3008.94507753983, 
3799, 2940, 3389.51332290472, 3090, 1701, 3363, 3033, 2325, 3941, 
3657, 3600, 3005, 4054, 3856, 3402, 2694.09822203382, 3413.03869100037
), tobacco01 = c(0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 
0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
0, 0, 1, 1), pscore = c(0.00988756408875347, 0.183983728674846, 
0.24538311074894, 0.170701594663405, 0.179337494008595,         0.0770304781540708, 
0.164003166666384, 0.0773042518100593, 0.0804603038634144,     0.0611822720382283, 
0.481204657069376, 0.166016137665693, 0.107882394783232,     0.149799473798458, 
0.04130366288307, 0.0360272679038012, 0.476513676221723, 0.214910849480014, 
0.0687582392973688, 0.317662260996216, 0.206183065905609,     0.336553699970873, 
0.0559863953956171, 0.103064791185442, 0.0445362319933672,     0.17097032928289, 
0.245898950803051, 0.146235179401833, 0.284345485401689,     0.152121397241563, 
0.0395696572471225, 0.116669642645446, 0.0672219220193578,     0.297173652687617, 
0.436771917147971, 0.0517299620576624, 0.140760280612358,     0.179726730598874, 
0.0118610298424373, 0.162996197785343), pscoreblocks = structure(c(1L, 
19L, 25L, 18L, 19L, 8L, 17L, 8L, 9L, 7L, 49L, 17L, 11L, 16L, 
5L, 4L, 49L, 22L, 7L, 33L, 21L, 35L, 6L, 11L, 5L, 18L, 25L, 15L, 
29L, 16L, 5L, 12L, 7L, 31L, 45L, 6L, 15L, 19L, 2L, 17L), .Label = c("    [3.88e-05,0.0099]", 
"(0.0099,0.0198]", "(0.0198,0.0296]", "(0.0296,0.0395]", "    (0.0395,0.0493]", 
"(0.0493,0.0592]", "(0.0592,0.069]", "(0.069,0.0789]", "(0.0789,0.0888]", 
"(0.0888,0.0986]", "(0.0986,0.108]", "(0.108,0.118]", "(0.118,0.128]", 
"(0.128,0.138]", "(0.138,0.148]", "(0.148,0.158]", "(0.158,0.168]", 
"(0.168,0.177]", "(0.177,0.187]", "(0.187,0.197]", "(0.197,0.207]", 
"(0.207,0.217]", "(0.217,0.227]", "(0.227,0.237]", "(0.237,0.246]", 
"(0.246,0.256]", "(0.256,0.266]", "(0.266,0.276]", "(0.276,0.286]", 
"(0.286,0.296]", "(0.296,0.306]", "(0.306,0.315]", "(0.315,0.325]", 
"(0.325,0.335]", "(0.335,0.345]", "(0.345,0.355]", "(0.355,0.365]", 
"(0.365,0.375]", "(0.375,0.384]", "(0.384,0.394]", "(0.394,0.404]", 
"(0.404,0.414]", "(0.414,0.424]", "(0.424,0.434]", "(0.434,0.444]", 
"(0.444,0.453]", "(0.453,0.463]", "(0.463,0.473]", "(0.473,0.483]", 
"(0.483,0.493]", "(0.493,0.503]", "(0.503,0.513]", "(0.513,0.522]", 
"(0.522,0.532]", "(0.532,0.542]", "(0.542,0.552]", "(0.552,0.562]", 
"(0.562,0.572]", "(0.572,0.582]", "(0.582,0.591]", "(0.591,0.601]", 
"(0.601,0.611]", "(0.611,0.621]", "(0.621,0.631]", "(0.631,0.641]", 
"(0.641,0.651]", "(0.651,0.66]", "(0.66,0.67]", "(0.67,0.68]", 
"(0.68,0.69]", "(0.69,0.7]", "(0.7,0.71]", "(0.71,0.72]", "(0.72,0.73]", 
"(0.73,0.739]", "(0.739,0.749]", "(0.749,0.759]", "(0.759,0.769]", 
"(0.769,0.779]", "(0.779,0.789]", "(0.789,0.799]", "(0.799,0.808]", 
"(0.808,0.818]", "(0.818,0.828]", "(0.828,0.838]", "(0.838,0.848]", 
"(0.848,0.858]", "(0.858,0.868]", "(0.868,0.877]", "(0.877,0.887]", 
"(0.887,0.897]", "(0.897,0.907]", "(0.907,0.917]", "(0.917,0.927]", 
"(0.927,0.937]", "(0.937,0.946]", "(0.946,0.956]", "(0.956,0.966]", 
"(0.966,0.976]", "(0.976,0.986]"), class = "factor"), blocknumber = c(1L, 
19L, 25L, 18L, 19L, 8L, 17L, 8L, 9L, 7L, 49L, 17L, 11L, 16L, 
5L, 4L, 49L, 22L, 7L, 33L, 21L, 35L, 6L, 11L, 5L, 18L, 25L, 15L, 
29L, 16L, 5L, 12L, 7L, 31L, 45L, 6L, 15L, 19L, 2L, 17L)), row.names =     c(NA, 
-40L), class = c("tbl_df", "tbl", "data.frame"))

Answer 1

他们给出了相同的结果，但我相信您是基于 position 而不是 name 的索引：

data %>% filter(blocknumber == 6) %>% apply_model()
# [1] -2090

如果我们尝试使用位置 6 索引 model_list，它不等于：

data_split <- data %>% group_split(blocknumber)
models <- data_split %>% map(apply_model)
models[[6]]
# [1] NA

但那是因为data_split[[6]] 与data %>% filter(blocknumber == 6) 不同：

data_split[[6]]
# # A tibble: 3 x 5
#   birthwt tobacco01 pscore pscoreblocks   blocknumber
#     <dbl>     <dbl>  <dbl> <fct>                <int>
# 1    4281         0 0.0612 (0.0592,0.069]           7
# 2    3459         0 0.0688 (0.0592,0.069]           7
# 3    3657         0 0.0672 (0.0592,0.069]           7

您可以通过分配名称然后按名称索引来解决此问题：

names(models) <- data_split %>% map("blocknumber") %>% map_chr(unique)
models[["6"]]
# [1] -2090

base::split 默认情况下也会保留名称，所以我通常更喜欢使用它：

models <- data %>% split(.$blocknumber) %>% map(apply_model) 
models[["6"]]
# [1] -2090